Laboratory of Plant Pathology, ORSTOM, BP A5, 98848 Noumea, New Caledonia

Background and objectives
Investigations on the pathology of Coffea arabica are being conducted in New Caledonia and in the South Pacific. They are aimed at gaining (1) an understanding of the overall functioning of the pathosystem which comprises coffee, its main pathogenic fungi (Hemileia vastatrix Berk. and Br., Colletotrichum gloeosporioides Penz. and Cercospora coffeicola Berk. and Cooke) and the environment, and (2) identifying the environmental conditions which affect the dynamics of these diseases. The ultimate objective of this research is to use modelling to devise a decision-making tool enabling the epidemic risk to be forecast in a defined environmental context.

Materials and methods
The experimental approach is based on monthly epidemiological observation carried out in small-holders' coffee plots (using specific pathological investigation methods) and on environmental characterization procedures. Such surveys were carried out in successive years. Management of these pathological and environmental data was achieved with the ORACLE database using queries that allow the focus to be varied and synoptic tables to be produced. Statistical interpretation is carried out with the ADE-4 (multivariate analysis and graphic expression of environmental data) software [1]. Within all the available multivariate analyses, the module STATICO (for STATIS and co-inertia), allowing a three-way analysis of a 'data cube' (variables X sites X observation dates) and a costructure analysis (by co-inertia) of two data cubes, was well adapted for such a spatial and temporal study [2].

Results and conclusions
In the annual cycles, surveys have confirmed the existence of a mosaic of highly diverse pathological situations. Using pathology or environment data, STATICO reveals an 'interstructure' ordering the sampling dates, then establishes a 'compromise' describing common structures at various dates and, finally, characterizes an 'intrastructure' giving, for each date, the divergences from the compromise structures. For the surveyed annual cycles, the locations of the study plots on the annual scatter diagrams vary slightly (while climatic conditions changed greatly), thus confirming the stability of coffee-pathogen relationships.

Co-inertia analyses between pathological and environmental data demonstrate links between them and reveal significant combinations of variables: (1) anthracnosis is linked to good soil structure, high pH values and low shade, (2) high pH and shade values and low altitude influence cercosporiosis, (3) rust is more likely to occur in sites featuring low soil pH values, low rainfall, high minimum temperatures, high shade and high altitude. This original correlation between edaphic factors and pathology is confirmed by considering results recorded over four successive survey years. Accordingly, it is may be assumed that the epidemiological diversity observed between sites is a reflection of edaphic diversity and is not influenced by the climatic variations. These major trends were used as a basis for modelling the infection risk for each disease and thus significant forecasts of the level of infestation were made for each plot. Modelling work is continuing to improve further this statistical approach by trying to forecast the kinetics of infection also.

To evaluate the variability caused by the diversity of the pathogens and the host, complementary studies are under way: (1) analyses of rust populations (RAPD and RFLP), (2) identification of rust races, (3) analyses of C. ;gloeosporioides populations (RAPD), and (4) genetic study on the homogeneity of surveyed plants (RAPD). The results of these biological investigations and the use of more focused subfiles of data, once included in the statistical model, should increase the accuracy of disease risk forecasts.

We assume that this synoptic and holistic approach, as well as the biometrical tools developed [3] may be of interest to other plant pathologists seeking to understand the functioning of their own pathosystems.

1. Thioulouse J, Chessel D, Doledec S, Olivier JM, 1997. Statistics and Computing 7, 75-83.
2. Simier M, Blanc L, Pellegrin F, Nandris D, 1998. Revue de Statistique Appliquée, in press.
3. Nandris D, Kohler F, Monimeau L, Pellegrin F, 1997. Proceedings 17th International Conference on Coffee Science, ASIC, Nairobi, Kenya, in press.